What ceph/ceph shipped

This week: cephsqlite supports concurrent readers; a ceph.alternate_name xattr for dentry alternate names; rgw metadata keeps layout logs from the stored bucket; mirror daemon rebalance fixed; MonClient and DaemonState fixes; doc updates for PG repair, FD Calc and librados object

This week in Ceph: cephsqlite now supports concurrent readers, and a new ceph.alternate_name extended attribute gets and sets the dentry alternate_name field. RGW metadata keeps the layout logs from the stored bucket, and the mirroring module fixes directory rebalance after a mirror daemon restart. On the manager side, DaemonState releases its lock before callbacks to avoid GIL inversion and parses wear_level metadata without atof ambiguity, and MonClient drops the ping connection before deleting its messenger. Documentation updates cover rewriting the inconsistent PG and pg repair sections, modernizing PG Calc with a new FD Calc page, and adding object size, I/O rate and omap guidance to the librados introduction.

This week: mgr keepalives fix daemon session faults, NFS adds clients_per_pool for CephFS exports, RGW rejects empty layout logs and temp URL copy headers, librbd relaxes exclusive lock for image-meta, and crimson OSD fixes EC hangs.

This week's changes include manager client keepalives so OSDs and MDSes no longer fault their mgr session at the idle timeout, and an optional clients_per_pool setting in the NFS manager module to spread exports across multiple CephFS client handles. RGW now rejects metadata puts that would leave bucket info layout logs empty and disallows temp URLs with x-copy-from headers. Librbd relaxes the exclusive lock requirement for image-meta set and rm, and Crimson OSD fixes three erasure-coded hangs tied to attr_cache. Documentation also gained cephfs-mirror failover steps and corrections across RADOS, cephadm and monitoring pages.

This week in Ceph: Objecter re-targets forced-OSD reads when a shard moves, ops apply their own pg_trim_to, MDS resets heartbeat in large cache walks, subvolume names are validated, and cephadm service discovery no longer needs the dashboard.

This week in Ceph: the Objecter now re-targets forced-OSD direct reads when the pinned shard leaves the acting set, and ops apply the freshly computed pg_trim_to in the transaction that computed it. The MDS resets its internal heartbeat during cache-wide walks in resolve and rejoin, and a dirfrag marked bad keeps projected_version in sync so subtree encoding no longer asserts. Subvolume commands reject empty and dot names that previously corrupted group directories, and subvolumegroup create supports namespace isolation. On crimson SeaStore the SPDK path is zero-copy, and ceph-mgr-cephadm now depends on routes so service discovery works without the dashboard. BlueStore reserves bdev label copies during online expansion.

This week in Ceph: fixes for Debian ceph-mgr restarts and missing modules after the module split, purged snaps and direct reads, plus new vector bucket policy management in RGW and a --protocols option for fs subvolume create.

This week in Ceph, a Debian packaging fix stops the per-module ceph-mgr packages from restarting the manager into start-limit-hit during upgrades, and also restores modules missing after the mgr module split. In RGW, the S3 vectors API gained vector bucket policy management, and the smb module now resolves RGW tenants automatically for tenant-prefixed buckets and users. The cephfs volumes module has a new --protocols option on fs subvolume create, with an smb preset, and the dashboard multi-site sync status card now offers a gateway toggle. On the storage side, the OSD no longer skips the purged snaps of the epoch after the last purged one, direct reads spanning two chunks are sent to both shards, and a pool created in degraded stretch mode keeps the right size and bucket target.

V19.2.7 is out. Install docs rebuilt for packages, tarballs, containers and mirrors. Manual deployment rewritten with checks after each daemon. MDS quarantine failures no longer leave clients blocked. Crimson Fast EC tests enabled and conditional writes on versioning-suspended bu

This week Ceph v19.2.7 was released. The install documentation was rebuilt across four pull requests, covering the Installing Ceph landing page, the manual install hub, the package, tarball and container image pages, an updated mirror list, a rewritten manual deployment procedure with checks after each daemon, and rewritten Windows install and configuration pages. In CephFS, the MDS now undoes the in-progress quarantine state when an operation fails and takes mds_lock before starting the internal quarantine request, so clients are no longer blocked by a failed quarantine. Crimson sees the remaining Fast EC tests enabled in test/librados/aio.cc, and osd seastore leaves boot mode before starting the background process. Fixes also landed for conditional writes on versioning-suspended buckets

Week 40: CephFS subvolume quarantine for security/recovery workflows, OSD test framework overhaul, automatic debug level guards to prevent production I/O storms, mirror sync fixes, and multisite fault injection tests.

Ceph week 40 highlights: CephFS gains subvolume quarantine for controlled access during recovery. OSD testing gets CRUSH-based placement and multi-zone EC support. New automatic debug level guards protect production from I/O floods. Mirror sync now correctly handles directory renames. RGW improves STS permission handling and adds multisite fault injection testing.

This week: cephx security hardening, Crimson Prometheus config support, RGW IAM policy API, centralized cephadm dependencies, and critical OSD peering fixes.

Ceph shipped significant improvements this week: hardened cephx service ticket verification for security, Crimson OSD now reads Prometheus port from config, dashboard gained RGW role policy management, and critical fixes for OSD write ordering and peering bugs. Service dependency logic is now centralized in cephadm for consistency.

Week 40: SimpleRADOSStriper fixes for AIO error handling and resource leaks, OSD replication improvements, upmap balancing corrections, and RocksDB v9.11.2 integration.

This week's Ceph updates focus on reliability and resource management. Key fixes include SimpleRADOSStriper properly capturing asynchronous I/O errors and preventing resource leaks, OSD replication handling balanced reads on replicas correctly, and upmap balancing improvements. Infrastructure changes include RocksDB dependency bump to v9.11.2 and daemon shutdown race condition fixes.